diff --git a/raylight/scripts/server.py b/raylight/scripts/server.py
index 5dbca78..a8e57b4 100644
--- a/raylight/scripts/server.py
+++ b/raylight/scripts/server.py
@@ -568,7 +568,12 @@ def _comfyui_alive() -> bool:
 
 @app.get("/health")
 def health():
+    # queued 统计真正 status=='queued' 的任务，不用 Q.qsize()——取消一条排队中的任务只是把
+    # JOBS 里那条的 status 标成 cancelled，不会把它从 Q 摘除，worker 要等真正 Q.get() 到它才
+    # 会跳过丢弃（见 worker() 里的注释）；qsize() 在那之前会把这些已取消的僵尸 entry 也算
+    # 进去，堆量取消时数字会明显偏高，跟 running 的统计口径（数 JOBS 状态）保持一致才准确。
     running = sum(1 for j in JOBS.values() if j.status == "running")
+    queued = sum(1 for j in JOBS.values() if j.status == "queued")
     comfy_alive = _comfyui_alive()
     ready = 1 if (STATE["model"] == "ready" and comfy_alive) else 0
     model_state = STATE["model"] if comfy_alive else "comfyui_unreachable"
@@ -578,14 +583,14 @@ def health():
         "gpu": "0,1,2,3",
         "partition": "raylight-fsdp",  # 不再是路由依据，纯展示——4卡是一个集群，没有单卡分区概念了
         "offload": "fsdp",
-        "queued": Q.qsize(),
+        "queued": queued,
         "running": running,
         "backend": BACKEND_URL,
     }
     return {
         "status": "ok" if ready else "unavailable",
         "ready": ready, "total": 1,
-        "queued": Q.qsize(), "running": running,
+        "queued": queued, "running": running,
         "backends": [backend],
         "gpus": _gpu_status(),
     }
diff --git a/raylight/server.py b/raylight/server.py
index 5dbca78..a8e57b4 100644
--- a/raylight/server.py
+++ b/raylight/server.py
@@ -568,7 +568,12 @@ def _comfyui_alive() -> bool:
 
 @app.get("/health")
 def health():
+    # queued 统计真正 status=='queued' 的任务，不用 Q.qsize()——取消一条排队中的任务只是把
+    # JOBS 里那条的 status 标成 cancelled，不会把它从 Q 摘除，worker 要等真正 Q.get() 到它才
+    # 会跳过丢弃（见 worker() 里的注释）；qsize() 在那之前会把这些已取消的僵尸 entry 也算
+    # 进去，堆量取消时数字会明显偏高，跟 running 的统计口径（数 JOBS 状态）保持一致才准确。
     running = sum(1 for j in JOBS.values() if j.status == "running")
+    queued = sum(1 for j in JOBS.values() if j.status == "queued")
     comfy_alive = _comfyui_alive()
     ready = 1 if (STATE["model"] == "ready" and comfy_alive) else 0
     model_state = STATE["model"] if comfy_alive else "comfyui_unreachable"
@@ -578,14 +583,14 @@ def health():
         "gpu": "0,1,2,3",
         "partition": "raylight-fsdp",  # 不再是路由依据，纯展示——4卡是一个集群，没有单卡分区概念了
         "offload": "fsdp",
-        "queued": Q.qsize(),
+        "queued": queued,
         "running": running,
         "backend": BACKEND_URL,
     }
     return {
         "status": "ok" if ready else "unavailable",
         "ready": ready, "total": 1,
-        "queued": Q.qsize(), "running": running,
+        "queued": queued, "running": running,
         "backends": [backend],
         "gpus": _gpu_status(),
     }